Machine state classification of electric track circuit by means of support vector machine

Abstract

The effectiveness of a track circuit monitoring system can significantly improve with the implementation of the automatic data
analysis. As part of this functionality, in our previous publication, we proposed the track circuit state classifier based on the logistic regression. However, this classifier has some limitations. In this article, we propose a more advanced classifier based on the support vector machine (SVM). We describe theoretical principles on which the classifier is built and demonstrate its work on synthetic rail circuit state data. We also show that the SVM track circuit state classifier with the Gaussian kernel requires fewer features than the classifier based on the logistic regression.